Fixed the problem where you couldn't run study inference on single video studies.

[Affan88888]

When running the demo with a study that contains only one DICOM/MP4, encode_study crashes with:
RuntimeError: Tensors must have same number of dimensions: got 2 and 1
This PR makes get_views (in the model.py file) always return a 2-D tensor of shape (N, 11), even when N=1, so concatenation with the video features (N, 512) works for any batch size.

get_views previously did:
stack_of_view_encodings = torch.stack([torch.nn.functional.one_hot(out_views,11)]).squeeze().to(self.device)

• F.one_hot(out_views, 11) → (N, 11)
• torch.stack([ ... ]) → (1, N, 11)
• .squeeze() removes all size-1 dims
  When N=1, (1, 1, 11) becomes (11,) (rank-1), which cannot be concatenated with (1, 512) along dim=1.

What’s changed:
Replaced the above with a shape-safe line that preserves the batch dimension and matches dtype/device:
stack_of_view_encodings = torch.nn.functional.one_hot(out_views, num_classes=11).float().to(self.device) # (N, 11)

• No stack / no squeeze → batch dim is retained for N=1.
• .float() → matches the feature tensor dtype for torch.cat.
• .to(self.device) → keeps tensors on the same device.

Expected behavior after this change:
The same code runs without error.
Shapes:

• features: torch.Size([1, 512])
• views: torch.Size([1, 11])
• concatenated: torch.Size([1, 523])

Backwards compatibility:

• For N>1, return shape remains (N, 11). No changes to callers.
• Only behavioral change is avoiding accidental rank-1 tensors for N=1.